Krishna Barri

• Built and run the analytics and AI platform for a med spa group of 15 locations plus corporate. Zenoti, NetSuite, Salesforce, Rippling, the ad platforms and the website all land in one warehouse, get modeled in dbt, and reach corporate leadership and location managers through a dashboard app I maintain. One place for revenue, cost, payroll and marketing instead of nine separate logins, over roughly 1.3 million invoice lines of history.
• Built product cost tracking that ties consumption to an effective-dated master price list, which gave the business per-service and per-provider margin for the first time. The POS does not report it.
• Reconciled collections at the invoice line, cash collected against what was billed, with the exceptions sorted into bundled fees, genuine double charges, and records the POS had orphaned upstream. Finance stopped chasing differences by hand.
• Built package and membership liability reporting on a prorated-paid basis net of refunds. Building it turned up three defects in the vendor's own liability report, which the business had been closing books against.
• Joined Meta ad spend and GA4 traffic to POS revenue so marketing sees return by campaign against collected dollars instead of platform-reported conversions. The first run found several thousand dollars a day still committed to campaigns that had stopped producing, and separately caught that the website had quietly stopped firing purchase events.
• Built the AI layer on that data. Staff ask about commissions, payroll, inventory, sales, memberships and executive KPIs in plain language and get the number the dashboard shows, create chart tiles by describing them, and receive scheduled reports in email or Teams. Every generated query runs under permission checks, a cost ceiling, and a refusal path for questions the data cannot answer.
• Split the AI work across specialized agents that own their own domain and hand off to each other, exposed over Model Context Protocol so approved AI tools query the warehouse under the same permissions a staff member has, not a shared service account.
• Score AI responses on correctness against known-good answers, latency, cost per question, and whether the answer was useful enough to act on, with token metering and a per-tenant spend cap so a runaway question cannot run up a bill.
• Rebuilt Zenoti ingestion after finding three defects that were quietly corrupting revenue, and got it reconciled to 99.87% against source. Documented five ways the vendor's appointment API returns incomplete data, including cancelled and no-show records it omits, and added reconciliation guards so appointment counts stop drifting.
• Nightly incremental refresh with deletion reconciliation, monitoring and failure alerts. The dashboard matches the POS the next morning.
• Stood the same platform up for a second organization on isolated per-tenant infrastructure, with onboarding automated down to a single provisioning command.

Tech: Python, FastAPI, SQL, dbt, PostgreSQL, AWS Redshift, S3, Glue, Lambda, EventBridge, SQS, API Gateway, ECS, Secrets Manager, Bedrock, Claude, SageMaker, Next.js, TypeScript, Tableau, Docker

• Led the automation and data migration group. Built a Python migration and validation tool that cut new-location onboarding time by about 30%.
• Moved client data off on-premise systems into AWS S3 and Redshift, with pipelines that could restart and roll back mid-run.
• Automated API and MSSQL extraction, SFTP retrieval and report validation across Redshift and MSSQL, which took a large recurring manual load off the delivery team.
• Wrote and tuned T-SQL and PL/SQL against the platform's transactional schema, the same schema I now pull from on the client side.
• Built Tableau dashboards with dynamic filters and KPIs over Redshift, MSSQL and the platform APIs, and ran Tableau Server: permissions, row-level security, content governance, performance tuning.

Tech: Python, .NET Core, SQL Server, SSIS, AWS S3, Redshift, Tableau, Tableau REST API

• Baxter (medical devices), Jun 2018 to Sep 2019: built ingestion and integration in Python and Java across APIs, S3 and Redshift, supported the Oracle JDE migration into the on-premise warehouse, containerized deployments with Docker and Kubernetes, and handled Basic, token and OAuth authentication against third-party platforms.
• Rhetorik, Mar 2017 to Jun 2018: REST services in Django REST Framework, Python ingestion of CSV feeds into MSSQL, and a hybrid framework covering API, UI and ETL automation.
• OM1 (healthcare outcomes data), Aug 2016 to Mar 2017: SQL for ETL validation and ad hoc analysis, Tableau dashboards for operational monitoring, and Spring Boot services capturing curator responses.